Repository navigation
Getting started
Open Claude Code in the repository folder and run this prompt:
read the CLAUDE.md file and README.md file and set up the environment
- Docker Desktop (or Docker Engine + Compose v2)
- Free host ports: 8110 (DCS console), 8111 (instructor station), 4840 (OPC UA), 5020 (Modbus TCP), 1880 (Node-RED); for the two other sites also 8120, 8121, 4841, 5021 (Port Arden) and 8130, 8131, 4842, 5022 (Vadhana)
git clone https://github.com/itotinsider/haltenberg-chemische-werke-uns.git
cd haltenberg-chemische-werke-uns
docker compose up -d --buildThen open the hub at http://localhost:8110. The plant starts at steady state; it doesn't need a start-up procedure.
On a Linux host (or WSL2 with the repository on the Linux file system) the DCS containers run as
user nobody (uid 65534) and must be able to write their data folders, data/ (Haltenberg),
data/port-arden/ and data/vadhana/. These folders come with the repository, but they belong to
you, so allow the DCS to write them once:
chmod a+rwx data data/port-arden data/vadhanaDocker Desktop on macOS and Windows doesn't need this. If a DCS container stops at start-up with
"Cannot write the data folder" (older versions: sqlite3.OperationalError: unable to open database file), this is the cause. A folder that was missing when the stack started was created by Docker
as root; then docker compose down, sudo chown -R $USER data, the chmod above, and start again.
| What | Where |
|---|---|
| Hub (all systems) | http://localhost:8110 |
| DCS operator console | http://localhost:8110/operator |
| Historian trends | http://localhost:8110/operator#trends |
| Detail display (e.g. the columns) | http://localhost:8110/operator#detail/C-401 |
| Trainers (overlays on the console) |
http://localhost:8110/operator?training=ot (IT/OT), training=eng, training=1
|
| Instructor station (fault scenarios) | http://localhost:8111 |
| OPC UA endpoint |
opc.tcp://localhost:4840/hcw/asu1/ (anonymous, no security) |
| Modbus TCP field I/O (raw) |
localhost:5020, unit 1 |
| Node-RED (OPC UA client flows) | http://localhost:1880 |
| 15-min averages (CSV, written by Node-RED) | exports/asu1_twa_15min.csv |
| Port Arden (US) hub · OPC UA |
http://localhost:8120 · opc.tcp://localhost:4841/PortArden/ASU7100
|
| Vadhana (IN) hub · OPC UA |
http://localhost:8130 · opc.tcp://localhost:4842/vadhana/asu2
|
The other two sites are described in Global sites.
Hub — one page linking every system, grouped by ISA-95 level.

New to plants? Read Plant automation in brief first.
- Open the hub at http://localhost:8110 and go to the DCS operator console.
- Switch on 🎓 Training and step through the process with ▶. Want more depth? Each step has "Go deeper" buttons, and the switch at the top of the card changes to the Engineering and IT/OT trainers without leaving the console.
- Click the value of
FIC-101(the air flow) to open its faceplate: PV, SP, OP, mode and a trend. - Click the column on the overview to open the C-401 detail display. Back and Close take you back.
- Open the instructor station (http://localhost:8111) and start GAN compressor K-503 motor trip.
- Back in the console, watch the Alarms tab (
I-503.TRIP), and in Trends the nitrogen header PIC-504 dip while the LIN backup FIC-508 takes over. - Recover: on the Interlocks tab reset I-503, start the GAN compressor and, once it runs, press Normal line-up. FIC-508 winds down as K-503 takes the header back.
docker compose down # stop
rm data/historian.db* data/dcs_state.json # wipe history and controller state (Haltenberg)
rm data/*/historian.db* data/*/dcs_state.json # ... and of the other two sites (keep the folders)
rm exports/*.csv # wipe the Node-RED averagesAfter pulling a change to dcs/config.py (controller tuning, SP limits), remove or rename
data/dcs_state.json before docker compose up -d --build. The checkpoint restores each loop's
tuning and limits, so otherwise the DCS keeps the old values.
After a change to the Modbus map (common/iomap.py), stop the stack first with
docker compose down, and only then docker compose up -d --build. Also move the checkpoint
aside. Otherwise the old DCS runs against the new plant for a few seconds and reads the wrong
registers. That's like a real I/O-map mismatch: run feedbacks look lost, the DCS trips the
machines, and the checkpoint saves those trips.
Start
Understand
Global DataOps
Use
Contribute